Open-source tools for the MolochAI API: one OpenAI-compatible endpoint for open-weight models with reduced refusal behavior.
Website · Documentation · API reference · Models · Contact
MolochAI is a developer API for open-weight language models. You send OpenAI-format requests to one base URL, choose a model by id, and pay with prepaid credit charged per request or a monthly plan with a token quota, under published limits. This organization holds the open-source tools for building on it: SDKs, a terminal client, a browser playground, runnable examples, the HTTP contract, and a mock server so you can develop and run CI without a key or quota.
Everything here is MIT-licensed.
| If you want to | Repository |
|---|---|
| Understand less-restricted models and choose how to run them | uncensored-ai |
| Call the API from Python, sync or async, with typed errors | uncensored-ai-python |
| Call the API from TypeScript or Node, with zero runtime dependencies | uncensored-ai-node |
| Chat with the models from a terminal | uncensored-ai-cli |
| Try models in a browser while the key stays on your server (also as a Docker image you build) | ai-model-playground |
| Copy a working example for a task, or for LangChain, LiteLLM or the Vercel AI SDK | uncensored-llm-examples |
| Integrate over plain HTTP, import the OpenAPI spec or Postman collection, or test offline | uncensored-ai-api |
| Measure time to first token, latency, tokens/s, error rates and cost of MolochAI or any other OpenAI-compatible API | llm-api-benchmark |
Create an API key in the dashboard under API keys. It is shown once.
export MOLOCHAI_API_KEY="your-api-key"
curl https://api.molochai.dev/api/v1/chat/completions \
-H "Authorization: Bearer $MOLOCHAI_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model": "glm-5.3-flash", "messages": [{"role": "user", "content": "Explain server-sent events in two sentences."}]}'With the Python SDK:
pip install "git+https://github.com/Moloch-AI/uncensored-ai-python@v0.1.2"from molochai import MolochAI
client = MolochAI() # reads MOLOCHAI_API_KEY
stream = client.chat.completions.create(
model="glm-5.3-flash",
messages=[{"role": "user", "content": "Explain server-sent events in two sentences."}],
stream=True,
)
for text in stream.text_deltas():
print(text, end="", flush=True)Already using an OpenAI SDK? Point it at https://api.molochai.dev/api/v1 with your MolochAI key, and read the compatibility notes for the parameters that are ignored or overridden.
| Where you work | What to use |
|---|---|
| Official OpenAI SDKs (Python, JavaScript) | Change the base URL and the model id. Examples cover streaming, tools, JSON output and every error shape. |
| LangChain, LiteLLM, Vercel AI SDK | Tested examples in uncensored-llm-examples, with the defaults each one gets wrong here (structured output, retries, stream errors). |
| Postman, Insomnia, Bruno, Hoppscotch | The Postman collection or the OpenAPI spec. |
| SillyTavern, Open WebUI, LibreChat, coding agents (Cline, Roo Code, Continue, Aider) | Setup guides on molochai.dev/integrations. |
your code ──► uncensored-ai-python · uncensored-ai-node · an OpenAI SDK · plain HTTP ─┐
terminal ──► uncensored-ai-cli (built on the Python SDK) ────────────────────────────┤
browser ──► ai-model-playground ──► its own server, which holds the key ───────────┤
▼
https://api.molochai.dev/api/v1
or, offline: the mock server from uncensored-ai-api (no key, no quota)
| Endpoint | What it does |
|---|---|
POST /chat/completions |
Chat completions, plain or streamed as server-sent events, with tool calling. Image input where the model and its server support it. |
GET /models |
Model ids, display names, and whether a server is serving each one right now. |
GET /api/auth/me |
Your plan, credit, and the context limit you get on each model. |
- Rate limit: 60 requests per minute per account, plus a 5-hour session cap and, on monthly plans, the plan's quota. See limits and errors.
- Not available: embeddings, image generation, audio, files, the Responses API, and JSON mode (
response_formatis ignored, so ask for JSON in the prompt and validate it). - Stream long replies. Plain requests that run past about 100 seconds can time out at the edge.
Checked on 2026-11-07. The live list is GET /models and molochai.dev/models. Facts about each base model come from its own model card.
| API id | Base model | Developer | Base license | Reasoning |
|---|---|---|---|---|
glm-5.3-flash |
GLM-5.3-Flash | Z.ai | MIT | On, high effort |
glm-4.7-flash-uncensored |
GLM-4.7-Flash | Z.ai | MIT | On, medium effort |
gemma4-26b-a4b-qat |
Gemma 4 26B A4B | Google DeepMind | Apache 2.0 | On, medium effort |
qwen3.8-27b-uncensored |
Qwen3.8-27B | Qwen Team | Apache 2.0 | On, medium effort |
llama-3.3-70b-instruct-abliterated |
Llama-3.3-70B-Instruct | Meta | Llama 3.3 Community License | Off |
These models are builds with reduced refusal behavior: they decline less often than the releases they are derived from. That does not mean they never decline, and it does not make them more accurate. The base model developers do not publish or endorse these builds. The Acceptable Use Policy applies, and an application that faces end users needs its own rules and moderation. The uncensored-ai guide explains the techniques and the trade-offs.
- Contributing: read the contributing guide, then the one in the repository you are changing. Issues and pull requests are welcome.
- Security: report vulnerabilities privately through the affected repository's Security tab. See the security policy.
- Help with your account, billing or the hosted API: molochai.dev/contact. Bugs in these tools go to the repository's issues. Never post an API key in an issue.
- Conduct: Code of Conduct.